onnxruntime
[CPU] Enable pre-packed weights sharing for MatMulNBits
#29163
Merged

[CPU] Enable pre-packed weights sharing for MatMulNBits #29163

derdeljan-msft
derdeljan-msft derdeljan-msft marked this pull request as ready for review 24 days ago
derdeljan-msft derdeljan-msft requested a review from jambayk jambayk 24 days ago
derdeljan-msft derdeljan-msft requested a review from tianleiwu tianleiwu 24 days ago
derdeljan-msft derdeljan-msft assigned derdeljan-msft derdeljan-msft 24 days ago
derdeljan-msft derdeljan-msft requested a review from yuslepukhin yuslepukhin 24 days ago
tianleiwu
tianleiwu commented on 2026-06-20
derdeljan-msft [CPU] Enable pre-packed weights sharing for MatMulNBits
64329127
derdeljan-msft [CPU] Fix MatMulNBits pre-packed weight sharing corrupting int8 buffers
84f3f9a3
derdeljan-msft Address PR comments
afb93d00
derdeljan-msft derdeljan-msft force pushed from 6da9ed91 to afb93d00 21 days ago
derdeljan-msft derdeljan-msft requested a review from tianleiwu tianleiwu 21 days ago
tianleiwu
tianleiwu commented on 2026-06-23
tianleiwu
tianleiwu commented on 2026-06-23
derdeljan-msft Make CPU initializers sharing MatMulNBits specific
a5947ef6
derdeljan-msft derdeljan-msft requested a review from tianleiwu tianleiwu 20 days ago
tianleiwu
tianleiwu dismissed these changes on 2026-06-24
derdeljan-msft Fix MatMulNBits LUT tests. Add coverage UT.
09025159
tianleiwu
derdeljan-msft Share MatMulNBits pre-packed weights cross-session via content tags
bc2a0e83
derdeljan-msft
tianleiwu tianleiwu requested a review from copilot-pull-request-reviewer copilot-pull-request-reviewer 18 days ago
tianleiwu
tianleiwu commented on 2026-06-25
tianleiwu
tianleiwu commented on 2026-06-25
copilot-pull-request-reviewer
copilot-pull-request-reviewer commented on 2026-06-25
tianleiwu tianleiwu dismissed their stale review 18 days ago
Superseded by re-review at bc2a0e8: the PR now uses the tagged-identity design and my earlier blocking concern is resolved. Remaining notes are non-blocking (left as COMMENT).
derdeljan-msft Key MatMulNBits prepack sharing by packed bytes; harden id and invariant
2f1e6ed8
github-actions
github-actions commented on 2026-06-25
derdeljan-msft Update onnxruntime/test/optimizer/dq_matmulnbits_fusion_test.cc
1bb7bd1c
derdeljan-msft debug: Try to disable newly introduced pre-packed weight sharing tests
926a6de0
derdeljan-msft Re-enable and reduce new tests
42b07b73
derdeljan-msft Reduce the number of tests ran in CI
a4b9a469
derdeljan-msft Restore full MatMulNBits prepack-sharing tests; keep --test_parallel …
14dd3a3a
derdeljan-msft Limit intra op thread pull for prepacking tests to 1
a074c775
derdeljan-msft derdeljan-msft requested a review from tianleiwu tianleiwu 18 days ago
derdeljan-msft derdeljan-msft enabled auto-merge (squash) 17 days ago
tianleiwu
tianleiwu approved these changes on 2026-06-26
derdeljan-msft derdeljan-msft merged a5d2663b into main 17 days ago
derdeljan-msft derdeljan-msft deleted the derdeljan/matmulnbits_cpu_weight_share branch 17 days ago

Login to write a write a comment.

Login via GitHub

Assignees
Labels
Milestone